Nvidia Launches Nemotron 3.5 Super VL 120B and Local Bolt Models
By Ali Al-Montaser — Senior AI Technical Editor
Nvidia has officially launched the open-weight Nemotron 3.5 Super VL with 120 billion parameters for enterprise agents, alongside two local Bolt models for AI PCs and DGX Spark workstations. Super VL and Bolt 74B-A9B are available now, with Bolt E4B arriving later this month. Nvidia claims Super VL outperforms Opus 5 and GPT 6.1 Sol on specialized agentic tasks, backed by enterprise adopters including Salesforce, Crowdstrike, and Perplexity.
Executive Overview
On Thursday, October 19, 2026, Nvidia officially announced the launch of new open-weight AI models: Nemotron 3.5 Super VL with 120 billion parameters for enterprise agents, and two local Bolt models for personal computers and workstations. Super VL and Bolt 74B-A9B are available immediately, while Bolt E4B will ship later this month. The models are positioned to run agentic workloads both in enterprise data centers and locally on AI PCs and DGX Spark workstations, with early adopters including Salesforce, Crowdstrike, and Perplexity.
📊 Official Technical Specifications & Data Sheet
| Technical Axis | Confirmed Official Data |
|---|---|
| 💰 Pricing & Usage Cost | Open-weight models are free to download and self-host; no token fees announced by Nvidia. Operating costs depend on private infrastructure (DGX Spark or Nvidia cloud). |
| 🌐 Platforms & Immediate Availability | Super VL and Bolt 74B-A9B are available now via official Nvidia channels; Bolt E4B ships later in October 2026. Bolt models run on AI PCs with RTX Spark chips and DGX Spark workstations through Microsoft and Dell. |
| ⚡ Performance & Speed Benchmarks | Nvidia claims Super VL completes specialized agentic tasks faster than Opus 5, GPT 6.1 Sol, Luna, and Astra (document intelligence, financial accounting, scientific experiments, health insurance claims, security attack resilience). |
| 🛡️ Security & Attack Resilience | Super VL includes security attack resilience as one of its post-training tasks; models are open and deployable within the enterprise's private infrastructure. |
| 🧠 Context Window | Context window size was not officially disclosed in the announcement; the model is open-weight and customizable. |
| 🌍 Arabic Language & Regional Support | No explicit Arabic support announced; the open-weight model allows for local fine-tuning on Arabic data. |
Deep-Dive Features & Architecture
Nemotron 3.5 Super VL features a 120-billion-parameter architecture. It is smaller than Nemotron 3 Ultra but achieves comparable accuracy in instruction-following and tool use, making it an ideal choice for enterprises seeking to balance capability with deployability. The model is fully open-weight, allowing organizations to fine-tune it on proprietary data and deploy it within their own data centers while retaining ownership of what they build—unlike closed frontier models.
On the local side, Nvidia launched two Bolt models: Nemotron 3.5 Bolt 74B-A9B 24GB, a 74-billion-parameter model designed for local agentic tasks that balances speed, multi-step reasoning, tool calling, and web browsing; and Bolt E4B, a smaller model aimed at conversational agents. These models run locally on AI PCs equipped with RTX Spark chips and on DGX Spark workstations, which are currently being sold through Microsoft and Dell, enabling users to delegate daily tasks without sending workloads to the cloud.
Nvidia confirmed that Super VL and Bolt 74B-A9B are available now, while Bolt E4B will be released later this month. Companies using Super VL include Skydance, EY, Ramp, Crowdstrike, Perplexity, and Salesforce, reinforcing the credibility of the performance claims.
Benchmark & Competitive Performance
According to Nvidia's announcement, Super VL outperforms Anthropic and OpenAI models—specifically Opus 5, GPT 6.1 Sol, Luna, and Astra—in task completion speed when fine-tuned on specific agentic tasks including document intelligence, financial accounting, scientific experiments, health insurance claims, and security attack resilience. While such claims are typical at major model launches, Nvidia backed them with a list of heavy-hitting enterprise customers actually adopting the model, giving the claims additional weight. Notably, these models compete with frontier AI labs that are also Nvidia customers and investors, at a time when those labs are turning to alternative chips such as AWS Trainium, Cerebras, and in-house silicon.
Industry Impact & Enterprise Adoption
The launch signals Nvidia's push to own the full stack of enterprise AI—from training chips to deployable open models. By offering open-weight models that can be fine-tuned and run on-premises, Nvidia lowers the barrier for enterprises concerned about data sovereignty and vendor lock-in. The local Bolt models extend this strategy to edge devices, enabling AI agents to operate directly on AI PCs and workstations without cloud dependency. This could accelerate adoption in regulated industries such as finance and healthcare, where data cannot leave the premises. The involvement of Salesforce, Crowdstrike, and Perplexity suggests strong enterprise interest in open-weight alternatives to closed APIs.
Conclusion
Nvidia's Nemotron 3.5 Super VL and Bolt models represent a significant step toward democratizing enterprise-grade AI agents. With open weights, local deployment options, and claimed performance advantages over leading closed models, Nvidia is positioning itself as a key enabler of the next wave of agentic AI. The immediate availability of Super VL and Bolt 74B-A9B, along with the upcoming Bolt E4B, gives developers and enterprises a powerful new toolkit for building and running AI agents securely and efficiently.
Media Source: TechCrunch AI | Fact Verification & Analysis: AI Tools Oasis
Frequently Asked Questions
Nemotron 3.5 Super VL has 120 billion parameters. It is an open-weight model designed for enterprise agents and delivers accuracy comparable to the larger Nemotron 3 Ultra in instruction-following and tool-use tasks.
Nvidia launched two Bolt models: Nemotron 3.5 Bolt 74B-A9B 24GB (74 billion parameters) for local agentic tasks, and Bolt E4B for conversational agents. Super VL and Bolt 74B-A9B are available now, while Bolt E4B will be released later this month (October 2026).
Bolt models run locally on AI PCs equipped with Nvidia RTX Spark chips and on DGX Spark workstations, which are being sold through partners such as Microsoft and Dell.
Nvidia claims Super VL completes tasks faster than Opus 5, GPT 6.1 Sol, Luna, and Astra when fine-tuned on specific agentic tasks including document intelligence, financial accounting, scientific experiments, health insurance claims, and security attack resilience.
The official announcement does not explicitly mention Arabic support, but the model is open-weight and can be fine-tuned on Arabic data, allowing Arabic developers to customize it locally without licensing restrictions.
AI Tools Oasis Team
Bringing you the latest news and analysis in the world of Artificial Intelligence with accuracy and credibility. Follow us for all updates.
